• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

4.5ghz w/o HT or Less then 4.5ghz w/ HT?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
but in all honesty, the most multitasking i ever do is opening more then 1 IE window at a time... I dont need to burn a disc and play a game at the same time.
 
If a CPU isn't at 100% you wouldn't notice a stutter anyway. Therefore when I said HT kicks in at 100%, in lamens terms you will only notice it at 100%. HT doesn't open a special portal that overclocks the CPU when it needs it or plucks some magic GHz from no where, it's just good at managing. It doesn't say process-A is taking up 100% cpu, let's magicly give process-B 25% cpu... instead it takes the 100% process-A and put's it at 75% top CPU usage allowing for process-B to use 25% cpu usage.

Thats the reason why they don't use HT when benchmarking.
 
Graphicism said:
If a CPU isn't at 100% you wouldn't notice a stutter anyway. Therefore when I said HT kicks in at 100%, in lamens terms you will only notice it at 100%. HT doesn't open a special portal that overclocks the CPU when it needs it or plucks some magic GHz from no where, it's just good at managing. It doesn't say process-A is taking up 100% cpu, let's magicly give process-B 25% cpu... instead it takes the 100% process-A and put's it at 75% top CPU usage allowing for process-B to use 25% cpu usage.

Thats the reason why they don't use HT when benchmarking.

Hyperthreading does nothing of the sort.

You cannot actually use a processor at the full 100% utilization; it's physically impossible to get ALL the execution units running at the exact same time. This applies to ALL modern pipelined processors; AMD's, Intel's, even NVIDIA's and ATI's video cores suffer from the same fate.

No matter what single task you're performing, there are going to be idle execution units somewhere. In a CPU, you'll have FPU units or MMX units or SSE units or 3DNow units or INT units just sitting twiddling their thumbs. In a video card, you'll have ROP's or Texture fetch units or Vertex processor units or something else also just sitting there waiting. It's more likely that most modern processors are only able to hit ~50-65% true utilization when considering all the "inactive" units, even when using the absolute most demanding software.

Hyperthreading is one method which allows more of those execution units to be active simultaneously, by way of processing two seperate process threads that are not identical. If one thread has all the FPU muxes in use, the INTeger or MMX or SSE units can be busy doing other things while waiting.

It does not do any sort of dynamic load balancing, and it has no relation to "total CPU utilization" (as shown by the Windows task manager). Hyperthreading works at 1% utilization or 99% utilization (again, as shown by the Windows task manager). If two threads can be processed in parallel (no matter the load) then the processor will run it that way.

The windows task manager is a little misleading when showing two processors in an HT system, because there really aren't two and neither one has a unique utilization capacity. The only reason a HT machine seems to run better at seemingly "maximum load" is simply because it really isn't -- when Windows task manager tells you 100%, it's not quite true. A hyperthreaded machine can still execute other functions outside of the "fully loaded" instruction queue. This is why it can still appear responsive in the face of a deadlocked process; there are still idle execution resources available that it can tap into.

It has nothing to do with hyperthreading "turning on" at 100%, nor does it have anything to do with "lowering the utilization" of any given thread. It is no more than a more advanced execution unit scheduler that can keep more things running simultaneously.

Whichi is also a good possibility as to why it's not stable with HT enabled at 4.5 ghz... More execution units running means more power consumption, more heat, more things making electrical noise and more chances for bad data to show up in an end result.
 
Last edited:
Vio1 said:
but in all honesty, the most multitasking i ever do is opening more then 1 IE window at a time... I dont need to burn a disc and play a game at the same time.


I was just using this as an example.

it all comes down to a matter of prefrences. if you dont do much multitasking you might have been better off going with a amd solution though.
 
I know how you feel Vio, even though 4.2 4.3 will have no visable effect on your games from 4.5, you want to run at your max overclock, :p im the same. Although ive found HT can help in games, well ive seen a few differences with it on and off. With HT off and you pause a game, some times the menu is very sluggish, i think this is more to do with my graphics card, but it never happend with HT, also coming out of games seems to be abit smoother. It really depends on if you mind a bit of 'lag', then again, with HT off you may not get any with a rig of that speed. I think best thing to do is just run without HT and then with, see what you prefer.


Edit, was it you who had a 2.8C m0 that done 3.6ghz with HT off? If so, seems odd you get two cpu's alike, not many people find anymore mhz with HT from what i can tell.
 
ya it was me with the 2.8c doing 3.6 without ht (long time ago).... All I know is that with HT enabled, my computer does some weird things.... with HT disabled, i have seen no ill effects of my oc.

Ive tried with both HT on and off at 4.5ghz, and to be honest ive noticed no difference (exept in benchies)....


I want to keep it at 4.5ghz cause 4.5 sounds so much cooler then 4.4 (or whatever would be stable with HT)....
 
what would you guys think is the max safe vcore with a vapochill? At what voltage am i just asking for a problem?
 
Vio1 said:
Anyone know the max safe voltage for pressies under vapochills?
1.6 is considered safe for the MOBO Vreg circutry. and even with that at your speed it will be very hot. Active cooling would be nice even if it's a fan over the mossfets. as far as the chip, dunno how high that should go. but mobo prolly die before the chip.
 
Graphicism said:
That's what I said, the only difference is I just don't sound like a spokes person for intel.
No, you really didn't. You said:
but HT is only enabled when the CPU hit's 100%...
Incorrect. It works at all levels of utilization.
but clearly it doesn't take effect until you hit 100% cpu right?
Incorrect. It works at all levels of utilization (again)
instead it takes the 100% process-A and put's it at 75% top CPU usage allowing for process-B to use 25% cpu usage.
Incorrect. It does not perform any sort of load balance, nor does it "deprioritize" individual threads.

I corrected all three of your errors and you're telling me you said the exact same thing? No, that too is incorrect. I just got done telling you how hyperthreading actually works, and it has nothing to do with being an "intel spokesperson". Hyperthreading is not something that Intel invented (well, except for the marketing term) and is not any sort of end-all be-all solution. It is one simplistic method of many available options for getting more execution units running simultaneously.

If you want to drag something fanboyish into the forums, do it elsewhere.
 
Ok, i have a new question: Is there a noticeble performance increase running a oc computer with memory at 1:1?


Currently my computer is overclocked to 16x281fsb = 4500mhz however im running my ram at 5:4 or 225mhz. Would I notice an increase in performance if I replaced my OCZ PC3500 EB with lets say PC4500 (or something that will reach 281mhz)

Worth my time and effort? If so, what ram would you recommend?
 
It mostly depends on timings. The break-even point is around 275-280mhz when comparing 5:4 ratio at tight timings versus 1:1 ratio at loose timings. If you can get higher than 280, you're usually going to have slightly better results at 1:1 if you can squeeze your ram into it.
 
The reason I'm changing my 3.0c at 3.7 is that it I have to disable HT to run it at that speed. I can't even run at 3.4 with HT enabled. HT is indispensable for me as I do a lot of video encoding in the background and without HT it makes the computer much harder to use.

I have done a few game tests running 4:5 and 1:1 and found nearly a 10% improvement in fps. This was running 200Mhz 2-2-2-5 1T & then 250Mhz 2-2-2-5-1T.
 
Vio1, very nice oc. I was just reading some old subscriptions and actually noticed something that might be very helpfull for you. You have the 3.2 Sl7b8 which can use the 14x multi with a new bios.

What I suggest is make sure you have HT enabled and see what you can do. I definetly notice it's benifets. Even if you have to knock it down to 4400 (boo-hoo, lol)

then use the 14x multi.

14*314fsb = 4400ghz, then keep the ram at 5/4 so it will be running 251mhz. If I am correct it should be able to handle 251mhz no prob.

Your fsb will jump 33mhz, and you will get 26mhz out of your ram. Not a bad increace at all.
 
ok, this is weird... in cpu-z, in the SPD tab lists my ram and SPD timing table.... it shows up as 2.5 4 3 9 (while my ram is 2.5 3 2 8).... why the difference? What is CPU z reading?
 
2.5 3 2 8 is that what it's rated for or is it set to that in the bios. Double check to make sure it's properly set in the bios.
 
Personally, I wouldn't turn HT off.

If I was given the choice, I think I'd trade HT for 1ghz more on my OC. But, we all know that's not going to happen.
 
Vio1 said:
Ok, i have a new question: Is there a noticeble performance increase running a oc computer with memory at 1:1?

Currently my computer is overclocked to 16x281fsb = 4500mhz however im running my ram at 5:4 or 225mhz. Would I notice an increase in performance if I replaced my OCZ PC3500 EB with lets say PC4500 (or something that will reach 281mhz)

Worth my time and effort? If so, what ram would you recommend?

I recommend tighter timings.
300 1:1 3-4-4-8 is generally way slower than 275 5:4 2-2-2-5
I say generally, because sandra will show you that the bandwidth is higher @ 1:1 vs 5:4, clock to clock.
SuperPi for example will perform better with tight timings.
 
Back